|Publication number||US7395252 B2|
|Application number||US 11/349,711|
|Publication date||Jul 1, 2008|
|Filing date||Feb 8, 2006|
|Priority date||Aug 26, 2003|
|Also published as||US20070094187, WO2005020044A1, WO2005020044A9|
|Publication number||11349711, 349711, US 7395252 B2, US 7395252B2, US-B2-7395252, US7395252 B2, US7395252B2|
|Inventors||Roger N. Anderson, Albert Boulanger|
|Original Assignee||The Trustees Of Columbia University In The City Of New York|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (8), Non-Patent Citations (2), Referenced by (19), Classifications (19), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is a continuation of International Application Ser. No. PCT/U504/028185, filed Aug. 26, 2004, published Mar. 3, 2005, which claims the benefit of United States provisional application No. 60/497,834 filed Aug. 26, 2003, each of which are incorporated by reference herein in their entireties, and from which priority is claimed.
This invention relates to systems and methods for supporting business decision-making in complex and uncertain process environments. More particularly, the invention relates analysis of probable process events and to the prediction of action outcomes to support business and engineering decision-making in the face of uncertainty.
A defining characteristic of the modern industrial society is the complexity of business processes that are involved in the production and delivery to market of almost every type of goods or services that are available today. Large or complex business processes are involved in the production and delivery of diverse products ranging, for example, from energy, health care, food, automobiles, sundry goods, telecommunications, music and other media. The business processes are complex not merely because of the physical size of the supply chain to market, but because of the complex array of decision-making variables that can affect production and delivery. For example, in the electric power industry, utility plant operating engineers and managers are faced with a complex array of decision making variables,—arising from deregulated markets, technology change, multiple weather events, physical failure situations and supply anomalies, and now the specter of terrorist attacks across multiple power grids. The variables have impacts of varying scale, e.g., local or global, short term or long term, on the business process. Further, the impact of each variable in real time may be dependent on the state of the other variables.
Conventional systems and methods for supporting decision-making deal with complexity in the business process by treating the business process in fragments. The business process is partitioned by organizational parts or divisions, and by hierarchal levels (e.g., regulatory control, supervisory control, and strategic planning). For example, regulatory control is used at a low level to tactically control local process variables. At the next higher level, supervisory control is used, for example, to optimize production schedules and to co-ordinate activities of different parts of the business process. Scheduling, operation planning, and capacity planning, or strategy functions, which may affect the business process on longer time scales, are carried out at even higher hierarchical levels. The business process decisions made at each level are often supported in isolation on the basis of ad hoc assumptions or static models of the process conditions at the other levels or of the state of other variables at that level. The fragmented approach to the complexity of the business processes can lead to gaps, and missed synergies or common mode interactions, which can affect the efficiency and security of the business process. While the fragmented approach for dealing with complexity may be adequate in static business environments, it does not exploit the potential for real time decision making support that is made possible by increasing investments in computerization and automation of the business processes.
Consideration is now being given to improving prior art systems and methods for business decision support. Attention is particularly directed to integrating supervisory and regulatory control as well as higher level strategy control for decision making under uncertainty in real time. Attention is also directed to integrating real option valuations in the decision making process.
The invention provides systems and methods for supporting business decisions under uncertainty. A stochastic controller system is used to optimize decision making through time. A unified approximate dynamic programming algorithm using reinforcement learning is implemented to treat multiple interconnected operational levels of a business process in a unified manner. The operational levels of business may, for example, include low levels such as supervisory and regulatory control levels and higher levels such as portfolio management, strategic (capacity) planning, operational planning, and scheduling.
Human decision making knowledge within all the operational levels of the business process can be captured in suitable learning matrices. The leaning matrices include expert knowledge on process situations, actions and outcomes at each level. The learning matrices are chained to obtain a unified representation of the entire business process. The learning matrices are then used as a source of “end games” knowledge (in the sense that computer chess games include a representation of expert play knowledge). A forward model of the business process is used to train the unified reinforcement learning algorithm to generate optimal actions at all levels of the business process. The endgame knowledge in the learning matrices is used to generate learning scenarios for the learning process. The stochastic controller system may include suitable process simulators to exercise the reinforcement learning in training.
The unified reinforcement learning algorithm uses suitable techniques to address the curse of dimensionality problem of dynamic programming algorithms The techniques used may include, for example, Codebooks (Vector Quantization), Neuro-Dynamic Programming (NDP), support vector machines, MAXQ hierarchical reinforcement learning and Lagrangian Decomposition. In one embodiment of the invention, NDP techniques approximate a cost-to-go function using neural networks for function approximation. The reduction in the demand for computational resources using the NDP techniques to approximate the cost-to-go function using neural networks for function approximation, can help the unified reinforcement-learning algorithm to scale with the dimensions of the business process. In another embodiment, support vector machines are used instead of neural networks for function approximation.
The unified reinforcement-learning algorithm may be configured to evaluate opportunities as real options. In one embodiment of the invention, the unified reinforcement learning algorithm processes is configured to generate actions or decisions that are always worth something if exercised immediately (that is they are “in-the-money” in options terminology), with respect to both financial profitability and engineering efficiency. Such in-the-money actions may be used to implement always advancing (non-losing) strategies/sequences of actions (that are called martingales using terminology from probability theory).
In an application of the present invention, a martingale stochastic controller is configured to carry out remote sub-sea decisions in a real-time affecting the form and timing of gas, oil, and water production in the ultra deepwater. An integrated reservoir asset and production model is developed. The model may include production constraints based, for example, on skin damage and water coning in wells. The stochastic controller is trained to generate flexible production/injection schedules that honor production constraints and produce exemplary field production shapes. The flexible production/injection schedules are optimized on the basis of total economic value increase (real option value+NPV) by the controller.
In another application of the present invention, the innervated stochastic controller is configured as a learning system in computer-based simulation and training tool—Decision Support Threat Simulator (DSTS), for training human operators to make optimal asset management decisions against threats or other exigencies in the business operations. The DSTS may be used for training power grid control operators to maintain the stability of transmission and distribution electric power grids, which are modeled with suitable simulators.
Further, the DSTS can be configured for human-in-the-loop control of the power grid as an adaptive aiding system, which is additionally useful for computer-based training of operators. The principal concepts of adaptive aiding systems are described, for example, in William, B. Rouse, Adaptive aiding for human/computer control, Human Factors, v.30 n. 4, p. 431-443, August 1988. In this configuration, the DSTS can be used to advise or guide operators on what actions to take in the same manner common car navigation systems are used to guide or direct car drivers. The DSTS suggests a control direction to the operators, while the operators operate the subject system being controlled, and further the DSTS also continually tracks the current state of the subject system to provide further control direction suggestions. The DSTS in its adaptive aiding configuration can be used either online while the operators are controlling actual grid operations, in a shadow mode, or offline for training exercises,
Further features of the invention, its nature, and various advantages will be more apparent from the following detailed description of the preferred embodiments and the accompanying drawings, wherein like reference characters represent like elements throughout, and in which:
The present disclosure provides control systems and methods for supporting business decisions under uncertainty. The business decisions may relate to actions at various levels of the operations in a business process, organization, entity or enterprise. The levels may include, for example, broad strategy levels at the upper ends of a business process and supervisory or regulatory control levels at the lower ends of the business process. The inventive control systems and methods may be used to optimize decision-making through time in the face of uncertainty. The decision-making may be based on a comprehensive consideration of all influencing factors, for example, economic as well as engineering factors, across each level of operations.
Suitable control links may allow rapid or real-time data communications between various levels or parts of a diverse business process, organization, entity or enterprise. The suitable control links may be implemented using, for example, electronic networks, wireless and other data communication technologies. The control links may include human input at one or more levels. Such implementations may advantageously allow business decision-making to be conducted in real time to actively respond to changing process conditions or outlooks.
In the present invention, a subject business process is viewed in a stochastic control framework. Under the stochastic control framework, business process outcomes or events are viewed as probable or stochastic events. The stochastic nature is due to uncertainties that come from both within the organization and external to it, such as markets, prices, risky internal development projects, etc.
An adaptable “Innervated” Stochastic Controller (ISC) is developed to optimize decision-making at all levels of the subject business process. (See
It will be understood that the qualifier ‘innervated’ is used herein in a sense analogous to the conventional dictionary meaning of the word: to supply or stimulate an organ or a body part with nerves. In a manner similar to the distribution of nerves in an animal, to and from its brain, spinal cord and all of its body parts, the ISC provides control links or “nerves” that extend through and connect various levels of a multi-level business process.
With reference to
Adaptive RL algorithm 102 uses reinforcement learning (dynamic programming) agent architecture to provide optimal decision-making choices under uncertain conditions. The reinforcement-learning has explicit goals (e.g., a fixed return, NPV, production success), can sense aspects of their environments, and can choose actions to influence their environments. The reinforcement-learning agents map business process situations to business process actions so as to, for example, maximize a numerical reward signal related to an explicit goal. The reinforcement learning agents deployed in RL 102 may be developed using any suitable reinforcement learning techniques that are commonly known, for example, in artificial intelligence, operations research, neural networks, and control systems. Descriptions of several common reinforcement learning techniques and agent architectures may be found, for example, in Richard S. Sutton and Andrew G. Barto, Reinforcement Learning: An Introduction, MIT Press, Cambridge, Mass., 1998.
To treat the various levels of a business process (like that shown in
Matrix representation 104 component of ISC 100 includes expert knowledge relating to possible situations, actions and outcomes of the business processes. This expert knowledge may be conveniently cataloged or stored in the form of learning matrices. These learning matrices may represent or correspond to a broad class of maps of the business process, which include, for example, maps of detections or observations to problems, and maps of problems to solutions.
Each learning matrix (e.g., matrix 500) can represent mapping rules for business processes. Further as known in the art, two matrices are needed to represent two layers of a neural network. Similarly, three matrices may be needed to represent three layers of a neural network. (See
The particular learning matrices that are included in matrix representation 104 component of ISC 100 may be combined or chained together to represent “common mode” interactions within different parts of the business process. The chained learning matrices can provide a universal approximate to represent the entire business process. For example, chaining a ‘detection to problem’ matrix with a ‘problems to solutions’ matrix generates a map for ‘detections to solutions’.
A business process can have an enormous number of inputs (dimensions). Conventional dynamic programming techniques that are useful for reinforcement learning can be computationally prohibitive because of the well known “curse of dimensionality.” In an embodiment of ISC 100 (e.g.,
ISC 100 may use suitable simulations or models (e.g.,
The training or reinforcement learning of ISC 100 may advantageously be compressed in time by the use of simulations of the business operations.
The inventive stochastic controller may be configured to optimize strategic business objectives and yet at the same time honor all necessary engineering or other physical constraints on business operations. For such configurations of the inventive stochastic controller, an integrated asset model of the subject business process is developed. The integrated asset model may be designed to predict all of the relevant quantities or parameters (e.g., economic, engineering, etc.) that are involved in malting asset management decisions at each level of operation in the business process. The integrated asset model may incorporate all of the necessary engineering or other constraints on the business operations so that the stochastic controller can be trained to avoid operations in undesirable or inaccessible variable space (e.g., the constrained engineering space).
With reference to
Trajectory 810 avoids the L-, bar-, and square-shaped barriers 840 while traversing the plant's state space. The shape of trajectory 810 is also determined by operational and business policies that may be incorporated in ISC controller 100. The incorporated policies define the actions to be chosen in every state visited by the system. The operational and business policies, which are incorporated in ISC controller 100, may relate to any quantifiable aspect of the business process including, for example, those governing the degree of acceptable risk or the level of expected returns from each action.
Financial institutions often have established policies to use options to hedge their investment portfolios against a spectrum of risk, among them interest rate risk, political risk, and market risk. Traditionally, the use of options has been limited to the finance industry (i.e. for financial instruments). The present invention advantageously enables the use of options to hedge risk in other business processes and operations. The reinforcement learning algorithm (102) of ISC 100 may be adapted to conform to a real options framework or methodology in which opportunity selections (i.e. the actions recommended ISC 100) are valued as options. Such options methodology provides decision—making flexibility in identifying profitable actions despite risks and uncertainties at each level of the subject business process controlled by ISC 100.
The reinforcement learning algorithm used in ISC 100 is particularly suited for tightly integrating real options valuations into RL algorithm 102. In dynamic programming, the problem of real option valuations and optimization can be treated as a stochastic control problem. The problem of optimizing real option value for the stochastic processes may, for example, be formulated as the problem of maximizing the expected value of discounted cash flows (DCF). The approximate dynamic programming technique for real option valuations gives the arbitrage-free price for an action/investment option when the given stochastic processes are constrained to be always in-the-money (i.e. a “martingale”) and the risk-free rate of return is used as the DCF discount factor. As reinforcement leaning is a type of approximate dynamic programming, the same reinforcement algorithm framework used in ISC 100 can be used to conduct real option valuation.
MSC system 1000 includes a reinforcement-learning controller 1002, a critic function 1003 that may be coupled to optional learning matrices 1004, and a model 1006 of plant 1200. Plant model 1006 may include temporal descriptions of the plant as a Plant State Vector, a Plant Action Vector, and a set of Plant System Measurements. The Plant State Vector may, for example, be a set of measures of the plant state (i.e. configuration) including information on plant component identification (ID) and types, and variable values for the plant components. The Plant State Vector is used as input for controller 1002. The Plant Action Vector is a set of values, messages, or instructions for placing the plant components into a desired state by each component. The Plant State Vector is used as input for controller 1002. The Plant System Measurements may be a set of measurements of plant condition and output, which can be used by the critic function to monitor plant performance. Model 1006 optionally also may include Configuration Tables, which is a set of named configurations of the plant.
The operation MSC system 1000 can be understood with reference to the learning process steps, which are shown as lettered steps a-d in
MSC system 1000 can be advantageously implemented to control complex and diverse business processes. MSC system 1000 makes it possible to implement real options as well as regulatory, supervisory and scheduling control, and operational and capacity planning functions that may be necessary to conduct technical and business operations in a businesses process. This capability of MSC 1000 is particularly advantageous for managing remote plant operations (e.g., the downhole “factory” notion in oil and gas extraction in ultra deepwaters).
In an exemplary implementation, MSC system 1000 is configured for ultra deepwater reservoir management applications in the oil and gas production industry. MSC system 1000 may be used to carry remote sub-sea decisions in a real-time affecting the form and timing of gas, oil, and water production in the ultra deepwater based on reinforcement learning. MSC system 1000 can achieve lower level control objectives at the same time that it maximizes the expected value of discounted cash flows under reservoir, technical, and market uncertainties.
For this exemplary implementation, the components of the subject plant 1200 (
A key to the reinforcement learning approach to the stochastic controller system for the ultra deepwater is the ability to simulate or model the reservoir to surface assets (e.g., oil field plumbing or distribution system). An integrated asset simulation capability may be provided for this purpose. Unlike common real option valuation methods, reinforcement learning by MSC system 1000 can be non-parametric. MSC system 1000 may be configured to directly sample the possible action paths via simulation instead of first building a parametric model of the stochastic variables (including financial variables) of the processes to be controlled. The controller learns from the experience gained by simulations of the business outcomes of flexible engineering decisions (
The Integrated Production Models (IPMs) that are used in MSC system 1000 may include IPMs that are known in the art. Simulation models and techniques that may be useful are described, for example, in John T. Han, “There is Value in Operational Flexibility: An Intelligent Well Application”, SPE Hydrocarbon Economics and Evaluation Symposium, Dallas, Tex., 5-8 Apr. 2003 (“SPE 82018”); S. H. Begg and R. B. Bratvold, and J. M. Campbell, “Improving Investment Decisions Using a Stochastic Integrated Asset Model”, SPE Annual Technical Conference and Exhibition, New Orleans, La., 30 Sep.-3 Oct. 2001 (“SPE 71414”); C. V. Chow, and M. C. Arnondin, “Managing Risks Using Integrated Production Models: Process Description”, Journal or Petroleum Technology, March 2000 (“SPE 57472”); and, Steve Begg, Reidar Bratvold, and John Campbell, “The Value of Flexibility in Managing Uncertainty in Oil and Gas Investments”, SPE Annual Technical Conference and Exhibition, San Antonio, Tex., 29 Sep.-2 Oct. 2002 (“SPE 77586”), all of which are incorporated by reference in their entireties herein.
The advantages of using real option valuations to manage assets by application of MSC system 1000 can be appreciated with reference to
The real time learning algorithm that is used in MSC system 1000 for reservoir management may be developed, and tested or verified using investigations or case studies of historical reservoir production data. In one such case study, production data on a particular deepwater turbidite field—South Timbaler block 295, is used. The South Timbaler block 295 field, which is the Gulf of Mexico's first deepwater turbidite field, has three major producing reservoirs. This field has been previously modeled using seismic imaging software. A reservoir model for South Timbaler block 295 developed for the “4D SeisRes” project is available to conduct such as case study. See e.g., Anderson, R. N., G. Guerin, W. He, A. Boulanger and U. Mello, “4-D Seismic reservoir simulation in a South Timbaler 295 turbidite reservoir”, The Leading Edge, 17(10),1416-1418, 1998. The case study investigation is designed to demonstrate that a suitable stochastic controller can maximize the expected value of discounted cash flows (i.e., net present value (NPV)+option value) while at the same time avoiding operational problems in oil field production. The case study can demonstrate that both an increase in value and a reduction of cost would have been achieved if the stochastic controller system MSC system 1000 had been used to select flexible production schedules in response to both production engineering problems and market price signals.
The case study can leverage the 4D SeisRes simulator work to construct a surrogate for a real time stochastic controller applicable to the e-Field real time reservoir management architecture (e.g.,
The integrated production modeling concepts described in
In the case study, various strategic production shaping objectives may be simulated, for example, Base-load gas, peaking oil production; Base-load oil, peaking gas production; Base-load oil, base-load gas production; and Peaking of both gas and oil production. A description of the notion of peaking and base load production shaping for fields and portfolios may be found, for example, in Roger Anderson and Albert Boulanger, “Flexible manufacturing techniques make ultra deep water attractive to independents”, Oil & Gas Journal, Aug. 25, 2003, which is hereby incorporated by reference in its entirety herein.
Exemplary flexible production options were generated from multiple reservoir simulation runs using a VIP Reservoir Simulator (Landmark) to estimate production engineering constraints or penalties. These flexible production options are shown in
The case study investigation may further simulate different water injector and production programs using actual prices for the 1990s, and evaluate the increase in option value for different strategic goals. The return-on-investment of these new production options for the field may be estimated, and compared with the historical program. The effectiveness of the stochastic controller to achieve real time production engineering objectives by simulating the policies derived from the controller may be studied.
For ease in understanding, the steps or tasks involved in the case study-based development of suitable stochastic controller for deepwater reservoir asset management are summarized below:
1. Develop a stochastic controller based on the approximate dynamic programming method of reinforcement learning. Various “curse of dimensionality” reduction methods may be explored including but not limited to VQRL, NDP, and MAXQ.
2. Assemble an integrated reservoir model based on fast 1D simulation and Integrated Production Management software.
3. Gather data set for South Timbaler block 295 and build simplified reservoir and integrated production model (IPM).
4. Establish production constraints based on skin damage, water coning, and breakthrough for selected wells in South Timbaler block 295 fields.
5. Train the stochastic controller to generate production/injection schedules that honor production constraints and produce exemplary field production shapes knowing the resultant price of oil and gas sales and water disposal expense from the 1990s well production history.
6. Derive the total economic value increase (real option value) based on these flexible schedules and the historical baseline and compare with the actual return-on-investment.
7. Identify cost reductions based on optimizing the schedule of engineering interventions in the subsurface via the stochastic controller's ability to minimize and avoid bad production/injection regimes such as skin damage and water coning and breakthrough.
The stochastic controllers of the present invention may be designed to control operations in fully automated business process environments (e.g., E-Field oil reservoirs, robotic manufacturing plants or assembly lines) with no or minimal human intervention. However, for some applications, the inventive stochastic controller may be specially configured as learning systems for operator training, for example, for business processes where manual intervention in operations is necessary or desirable.
The Learning System 1400 may be used to train power control system operators, for example, to take suitable actions to respond quickly to fluctuations or interruptions in electricity flow anywhere in the power grid. Learning System 1400 may be configured as a computer-based simulation and training tool (e.g., Decision Support Threat Simulator (DSTS) 1500,
Learning system 1400 also may be configured to act on its own and take automatic control actions, for example, in response to fast moving events that cannot be quickly or properly responded to manually by human operators. In this configuration, the power grid is subject to mixed automatic and human control. The shifting automatic-manual responsibility for control actions is tracked and learned by the reinforcement learning system and the models it uses.
Using suitable simulation models of the subject power grids, the DSTS can link and analyze specific threat events (e.g., storm outages, normal failure, natural disaster, man-made threats) on power grid 1600, and generate planned and prioritized responses to the specific threat events by analysis of catastrophic grid-wide sequences that the learning system computer has seen or learnt before. The planned responses to specific events may be designed to meet safety, reliability, engineering, security, and financial objectives conjointly.
The learning system computer automatically and continually “learns” as it absorbs, stores, and makes relationships among discrete simulation runs of the power grids. Grid operation intelligence and expertise is extracted and embedded in DSTS 1500 by the use of learning matrices so that optimal learned responses to crises and attacks anywhere on the integrated grid are readily available. Thus, operators using DSTS can be trained and prepared to respond to unforeseen contingencies without the need to run new simulations at the moment. DSTS can identify and optimize answers to the following problems throughout the organization that impact the stability of the grid:
In one implementation of learning system 1400 as a computer-based simulation and training tool (e.g., DSTS 1500 (
DEW also may be used to model regional grid behavior using, for example, Federal Energy Regulatory Commission (FERC) data. DEW models the power flows in the grid, plans peak load leveling, peak shaving, voltage correction, right sizing of equipment, and automated data modeling.
In DSTS 1500, learning system 1400 and DEW simulators 1550 are linked into an integrated software tool. The integrated software tool may be configured, for example, to identify new threats to the grid, pinpoint new failure points in the grid, identify new downward propagation patterns or domino effects, prioritize and plan the restoration sequence in response to a disturbance in the grid; estimate and reduce restoration costs, identify acceptable solutions quickly, select most efficient economic options, eliminate delays in transmitting vital data, and facilitate getting the right data to and from the right people. DSTS 1500 provides a way to eliminate the “wish I could have seen it coming” reaction by making failure models predictive and storing responses and remediation scenarios for grid failures in the future.
Reinforcement learning controller 1002 in learning system 1400 may be implemented using neuro-dynamic programming (NDP) approaches or support vector machines to deal with computationally expensive cost-to-go function evaluations. Expert knowledge for learning matrices 1004 (e.g., similar to
Furthermore it is possible to use DSTS for threats to infrastructures that are not directly related to power such as ports and transportation hubs or transportation networks if simulations are available. An example of such a simulation is the port simulator, PORTSIM developed by the Decision and Information Science Division, Argonne National Laboratory. Another example is the transportation simulator TRANSIM, developed by the D4 division, Los Alamos National Laboratory and marketed by IBM Business Consulting.
In accordance with the present invention, software (i.e., instructions) for implementing the aforementioned innervated stochastic controllers and systems can be provided on computer-readable media. It will be appreciated that each of the steps (described above in accordance with this invention), and any combination of these steps, can be implemented by computer program instructions. These computer program instructions can be loaded onto a computer or other programmable apparatus to produce a machine, such that the instructions, which execute on the computer or other programmable apparatus create means for implementing the functions of the aforementioned innervated stochastic controllers and systems. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the functions of the aforementioned innervated stochastic controllers and systems. The computer program instructions can also be loaded onto a computer or other programmable apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions of the aforementioned innervated stochastic controllers and systems. It will also be understood that the computer-readable media on which instructions for implementing the aforementioned innervated stochastic controllers and systems are be provided, include without limitation, firmware, microcontrollers, microprocessors, integrated circuits, ASICS, and other available media.
It will be understood, further, that the foregoing is only illustrative of the principles of the invention, and that various modifications can be made by those skilled in the art, without departing from the scope and spirit of the invention, which is limited only by the claims that follow.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5625751||Aug 30, 1994||Apr 29, 1997||Electric Power Research Institute||Neural network for contingency ranking dynamic security indices for use under fault conditions in a power distribution system|
|US5875431 *||Mar 15, 1996||Feb 23, 1999||Heckman; Frank||Legal strategic analysis planning and evaluation control system and method|
|US6154731 *||Sep 3, 1998||Nov 28, 2000||Monks; Robert A. G.||Computer assisted and/or implemented process and architecture for simulating, determining and/or ranking and/or indexing effective corporate governance using complexity theory and agency-based modeling|
|US6169981 *||Jun 4, 1997||Jan 2, 2001||Paul J. Werbos||3-brain architecture for an intelligent decision and control system|
|US6308162 *||May 21, 1998||Oct 23, 2001||Khimetrics, Inc.||Method for controlled optimization of enterprise planning models|
|US6311144 *||Jul 31, 1998||Oct 30, 2001||Nabil A. Abu El Ata||Method and apparatus for designing and analyzing information systems using multi-layer mathematical models|
|US6321205 *||Aug 17, 1998||Nov 20, 2001||Value Miner, Inc.||Method of and system for modeling and analyzing business improvement programs|
|US6807537 *||Dec 4, 1997||Oct 19, 2004||Microsoft Corporation||Mixtures of Bayesian networks|
|1||Shervais, S., "Adaptive Critic Based Adaption Of A Fuzzy Policy Manager For A Logistic System," IEEE 0-7803-7078-3/01, 2001.|
|2||*||Stephen Shervais et al., Adaptive Critic based Adaptation of a Fuzzy Policy Manager for a Logistic System, 2001, IEEE, 0-7803-7078, 568-573.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7865384 *||Aug 17, 2007||Jan 4, 2011||International Business Machines Corporation||System and method for identifying and reducing costs of information technology actions in real time|
|US8014992 *||May 31, 2007||Sep 6, 2011||William H Smith||Electric power system training simulation software and process for electric power system training|
|US8265874||Apr 21, 2010||Sep 11, 2012||Saudi Arabian Oil Company||Expert system for selecting fit-for-purpose technologies and wells for reservoir saturation monitoring|
|US8412501 *||Jun 16, 2010||Apr 2, 2013||Foroil||Production simulator for simulating a mature hydrocarbon field|
|US8421469 *||Oct 22, 2010||Apr 16, 2013||Tesla Motors, Inc.||Method and apparatus for electrically cycling a battery cell to simulate an internal short|
|US8457997 *||Aug 30, 2005||Jun 4, 2013||Landmark Graphics Corporation||Optimization of decisions regarding multiple assets in the presence of various underlying uncertainties|
|US8468041 *||Dec 22, 2004||Jun 18, 2013||Oracle America, Inc.||Using reinforcement learning to facilitate dynamic resource allocation|
|US8560476||Jan 24, 2008||Oct 15, 2013||The Trustees Of Columbia University In The City Of New York||Martingale control of production for optimal profitability of oil and gas fields|
|US8725481||Apr 3, 2013||May 13, 2014||Exxonmobil Upstream Research Company||Parallel adaptive data partitioning on a reservoir simulation using an unstructured grid|
|US8725625||Oct 8, 2012||May 13, 2014||The Trustees Of Columbia University In The City Of New York||Capital asset planning system|
|US8725665||Aug 20, 2012||May 13, 2014||The Trustees Of Columbia University In The City Of New York||Metrics monitoring and financial validation system (M2FVS) for tracking performance of capital, operations, and maintenance investments to an infrastructure|
|US8751421||Jan 15, 2013||Jun 10, 2014||The Trustees Of Columbia University In The City Of New York||Machine learning for power grid|
|US8856048||Oct 15, 2009||Oct 7, 2014||International Business Machines Corporation||Method, system, and computer program product for automatically applying a predictive temporal profile to computer resource management decisions|
|US8972066||Sep 20, 2010||Mar 3, 2015||The Trustees Of Columbia University In The City Of New York||Decision support control centers|
|US9026417||Oct 20, 2008||May 5, 2015||Exxonmobil Upstream Research Company||Iterative reservoir surveillance|
|US9031821||Nov 13, 2012||May 12, 2015||Foroil||Production simulator for simulating a mature hydrocarbon field|
|US20060247990 *||Aug 30, 2005||Nov 2, 2006||Keshav Narayanan||Optimization of decisions regarding multiple assets in the presence of various underlying uncertainties|
|US20110313744 *||Dec 22, 2011||Foroil||Production Simulator for Simulating a Mature Hydrocarbon Field|
|US20120013341 *||Jan 19, 2012||Tesla Motors, Inc.||Method and Apparatus for Electrically Cycling a Battery Cell to Simulate an Internal Short|
|U.S. Classification||706/45, 706/52, 706/46|
|International Classification||G06F7/00, G06Q10/00, G06F3/00, G06F3/02, G06F15/18, G06F1/00, G06N5/00, G06F17/00|
|Cooperative Classification||G06N5/022, G06N99/005, G06Q10/04, G06Q10/10|
|European Classification||G06N99/00L, G06Q10/10, G06Q10/04, G06N5/02K|